International Journal of Transportation Science and Technology 5 (2016) 257–267

Contents lists available at ScienceDirect

International Journal of Transportation

Science and Technology
journal homepage: www.elsevier.com/locate/ijtst

Methodology for estimating capacity and vehicle delays

at unsignalized multimodal intersections
S. Ilgin Guler a,⇑, Monica Menendez b
a
Department of Civil and Environmental Engineering, Penn State University, USA
b
Institute for Transport Planning and Systems, ETH Zurich, Switzerland

a r t i c l e i n f o a b s t r a c t

Article history: An uncontrolled intersection is one where neither traffic signs nor lights, are used to guide
Received 20 September 2016 the traffic through the intersection, but instead the drivers decide on the right of way based
Received in revised form 28 February 2017 on some standard priority rules. This paper establishes a methodology to systematically
Accepted 2 March 2017
evaluate the expected average delays at multi-modal uncontrolled intersections. The
Available online 11 March 2017
methodology considers the demand values of different traffic streams, along with the pri-
ority and direction of each stream to determine the capacity available for each. Using this
Keywords:
capacity and the formula for delay adapted from the highway capacity manual, the average
Multimodal
Intersection
delay for each vehicle stream can be determined. The methodology is tested using data col-
Uncontrolled lected at five locations in Zurich, Switzerland. The results show that the methodology can
Capacity predict the delay of different vehicle streams to within 4 s/veh, and also can identify the
Delay streams for which large delays would be expected.
Ó 2017 Tongji University and Tongji University Press. Publishing Services by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).

Introduction

An uncontrolled intersection is one where neither traffic signs nor lights are used to guide the traffic through the inter-
section, but the drivers decide on the right of way. To do so, they typically follow clear rules, such as public transportation
having priority over cars, or cars inside a roundabout having priority over cars outside the roundabout.
Many of this type of intersections, including ones with multiple modes and relatively large demands can be found in Eur-
ope. However, without a systematic and practical method to assess their operation it is hard to determine whether an uncon-
trolled intersection would be able to satisfy the traffic demands with an acceptable level of service. Therefore, it is important
to develop an analytical tool for estimating capacities and the resulting vehicle delays in the presence of multimodal streams,
which may be encountered at uncontrolled intersections.
Two general approaches for determining capacity at uncontrolled intersections exist. The first approach utilizes gap
acceptance concepts to determine the capacity of a low priority stream. This implies that a distribution of headways among
vehicles in the high priority stream is assumed, and drivers trying to cross it accept or reject these gaps according to certain
criteria. Some of the initial research on utilizing gap acceptance concepts to determine capacity at intersections was carried
out by Harders (1968). Later research has shown how to use multiple variations of these concepts to analyze different prob-
lems such as: uncontrolled intersections (Tanner, 1967; Highway Capacity Manual, 2010), limited priority merges where

Peer review under responsibility of Tongji University and Tongji University Press.
⇑ Corresponding author.
E-mail address: iguler@engr.psu.edu (S.I. Guler).

http://dx.doi.org/10.1016/j.ijtst.2017.03.002
2046-0430/Ó 2017 Tongji University and Tongji University Press. Publishing Services by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
258 S.I. Guler, M. Menendez / International Journal of Transportation Science and Technology 5 (2016) 257–267

major stream cars either allow minor stream cars to enter, or the minor stream cars push their way in even if they don’t have
the right-of-way (Ma et al., 2013; Menendez et al., 2015; Troutbeck, 1999), and highway merges (Bunker and Troutbeck,
2003; Troutbeck and Kako, 1999; Menendez, 2006). However, these papers do not present a systematic methodology for
investigating multimodal uncontrolled intersections.
The second approach uses conflict theory, which was first introduced by Gleue (1972) for signalized intersections. The
goal is to establish a conflict matrix for each traffic stream and calculate the additive conflict flows (ACF). The conflict matrix
establishes the different streams that are in conflict with each other and could affect each other’s capacity. Then, the main
idea is that depending on the conflicting streams two factors can lead to larger capacity for each traffic stream: (1) the prob-
ability that the road is not blocked, and (2) the probability that the gap on the main stream is large enough. The ACF approach
has been used to model all-way-stop-controlled intersections (Li et al., 2009; Wu, 2000), two-way-stop-controlled intersec-
tions (Brilon and Wu, 2001; Li and Deng, 2008; Li et al., 2009) also with multiple-lane approaches (Li et al., 2011), uncon-
trolled intersections (Li et al., 2009), and roundabouts (Qu et al., 2014). Multiple modes along with cars, such as
pedestrians and bicycles can also be included in the analysis of capacity with this method (Brilon and Miltner, 2005; Li
et al., 2009). However, the conflict theory approach has a pre-defined set of possible conflicts that are considered, and does
not systematically consider different traffic streams. This bounds the analysis to the intersection configurations presented in
those papers.
The goal of this paper is to develop a systematic methodology for determining capacities available for multimodal
streams at uncontrolled intersections, which can be used for delay estimation of the vehicle streams. The methodology pre-
sented in this paper is systematic and easy to adapt to the local conditions if necessary. It is also tested with real data from
five locations in Zurich, Switzerland.
The rest of the paper is organized as follows. First, an overview of the background research which forms the foundations
of this work is provided. Second, the methodology employed to systematically analyze the capacities and resulting delays at
uncontrolled intersections with multiple modes is described in detail. Next, the methodology is empirically evaluated using
data from five locations in Zurich. Finally, some concluding remarks are presented.

Background

This work uses the principles of ACF, but develops a systematic methodology to determine the equivalent of a conflict
matrix at complex uncontrolled intersections with multiple modes, and then the resulting capacity of each traffic stream.
The results are systematically obtained following two consecutive iterative algorithms. The advantage of the methodology
developed in this paper is that it can be applied to many different intersections with different geometrical and operational
properties, and due to its systematic nature is programmable. This work builds on the work of Pitzinger and Spacek (2009)
which identified a methodology for assessing delays at uncontrolled intersections. The methodology presented in Pitzinger
and Spacek (2009) relies on some subjective judgements, which creates additional complications in the calculations. This
work expands on Pitzinger and Spacek (2009) by developing a simpler, more systematic and comprehensive approach. A
clear methodology is developed that does not require any judgement and hence can be programmed to quickly analyze mul-
timodal intersections.
The first step of this work was to empirically determine the saturation flows (S) of different modes, with a special focus on
car traffic, and how the other modes affect car traffic. To do so, 43 h of video data were collected across 20 intersections
throughout Switzerland to estimate standard values for the saturation flow of different transport modes (cars, buses, trams,
and pedestrians) at uncontrolled intersections. Notice that these values are slightly lower than those that would be observed
at controlled intersections, as drivers typically behave more carefully in the total absence of traffic signs or signals. The
results were used as default values in the subsequent calculations, and are provided in Table 1.
In the case of cars, two values were observed, one when cars enjoy the highest priority, and one when they do not. The
latter is smaller than the former, as in this case cars tend to be even more careful when crossing the intersection. In the case
of pedestrians, S represents the flow of pedestrians needed to completely block car traffic. For the pedestrian calculations, S
was found to depend on q, the average number of pedestrians crossing the intersection simultaneously. This parameter q can
take on the values 1 through 5, and if the average number of pedestrians passing together is greater than 5, q = 5. For the case
of trams and buses, S was observed based on the actual operating constraints of these vehicles (e.g., minimum headways
required). More details can be found in Menendez et al. (2015).

Table 1
Saturation flow values of different modes.

Mode Saturation flows (S)

Car Highest priority 1750 veh/h
Not highest priority 1650 veh/h
Pedestrian 900q ped/h where q is the average number of
pedestrians crossing the intersection together q = [1–5]
Trams 340 trams/h
Bus 600 buses/h
 S.I. Guler, M. Menendez / International Journal of Transportation Science and Technology 5 (2016) 257–267 259

Next, intersections where only two streams interact were analyzed. For this, more empirical data were collected at 10
intersections for a total of over 5 h at various sites within Zurich. Different combinations of two stream interactions were
considered: Car/Car, Car/Pedestrian, Car/Bus, and Car/Tram. The data was used to verify (or modify where appropriate)
the originally proposed methodology (Pitzinger and Spacek, 2009) to determine the conflict flows of each stream. The main
goal was to determine an ‘‘effective capacity” for each traffic stream based on its own demand, the demand of the conflicting
traffic stream, and the saturation flows from Table 1. More on this is discussed in Section 3. Results show that with some
modifications to the original methodology, traffic operations at two stream intersections can be predicted within an average
error of 1.9 s/veh. Again, more details on this can be found in Menendez et al. (2015).
The following section describes the methodology developed to determine the effective capacity of different traffic streams
at an uncontrolled intersection. This can in turn be used to calculate the average delay expected for each. This methodology
builds on the findings of the two-stream interactions, with the assumption that larger intersections with multiple streams
can be broken down into the interactions between pairs of streams and the effects can be combined.

Methodology

The goal of this section is to develop a methodology to systematically analyze complex intersections, where multiple
interactions take place at the same location. The methodology applies to intersections that could potentially discharge at
capacity (i.e., traffic operations are isolated and are not affected by other disruptions, such as queues from a downstream
location), however remain under saturated over the long run.
To analyze any complex intersection, it first needs to be broken down into its different components. This means that: (i)
all the different modes and their corresponding priority levels need to be identified; and (ii) the direction of each mode needs
to be identified. The first step implies that there are clear rules on the different modes regarding which have priority over
each other. For example, in Swiss cities trams always have priority over cars, as they cannot easily accelerate and decelerate.
These rules can vary among different locations. It is also possible that multiple traffic streams have the same priority. The
second step implies that the interactions between all traffic streams could be broken down as conflicting (C) or non-
conflicting (NC). Two NC (non-conflicting) streams would not directly interact with each other nor compete for space at
the intersection (for example two streams running parallel to each other), but could take advantage of each other’s priority.
Two C (conflicting) streams would compete for the same space on the roadway at the intersection. An example to better
illustrate the definitions used in this paper is provided in Fig. 1. In this figure, the tram and car streams are NC, and the pedes-
trian stream is C with both, the car and the tram streams. Also, the trams have the highest priority, which is indicated by the
number 1 within parenthesis, followed by pedestrians which have 2nd priority, and cars which have 3rd priority (according
to Swiss rules). Again, in the case where multiple streams have the same priority, they would be assigned the same number
within parenthesis.
Notice that intersections can have any shape (i.e., they do not necessarily have to be crossed intersections). Nevertheless,
all streams must be abstracted into parallel or perpendicular streams (i.e., NC or C). Once this abstraction of an intersection is
made, the general methodology to analyze a complex intersection follows four steps:

1. Determine general form of the effective capacity equation based on the type of interaction(s).
2. Calculate inputs for the effective capacity equation by analyzing the conflicting traffic stream(s).
3. Compute the final effective capacity of each stream; and
4. Calculate delay.

These four steps are described in detail below. They must be carried out sequentially for each traffic stream of interest j.
We suggest to start with the stream with the highest priority (tram in the case of Fig. 1), continue with the second highest,
etc. They lead, at the end, to the calculation of average delay per vehicle for each traffic stream, and not for the whole
intersection.

Step 1: Determining the general form of the effective capacity equation

The general form of the effective capacity equation can be determined by following the general algorithm shown in Fig. 2.
The main idea is to determine how C or NC streams may affect the capacity of each other.

Pedestrian (2)

Tram (1)

Car (3)

Fig. 1. Illustration for definitions of priority, and conflicting (C) vs. non-conflicting (NC) traffic streams.
260 S.I. Guler, M. Menendez / International Journal of Transportation Science and Technology 5 (2016) 257–267

Fig. 2. Flowchart for determining the general form of the capacity equation.

The first possibility is that the capacity of the stream of interest, j, is either (i) not affected by any C stream, or (ii) affected
by some C stream and cannot take advantage of any NC stream. Condition (i) means that j either does not have any C stream,
or it has priority over all C streams it has. Condition (ii) means that at least one C stream has higher priority than j and all its
NC streams. In this case, the effective capacity of the given stream, Lj , is calculated as in Eq. (1).

Lj ¼ Sj  bj ð1Þ

where Sj is the saturation flow for the given stream, and bj is a factor which modifies the capacity based on the flow ratios of
the C streams (more on b is discussed in Step 2). An example of this would be the tram or pedestrian streams in Fig. 1.
The second possibility is that the traffic stream is affected by C streams, but can also take advantage of at least one NC
stream which is not interrupted by the C stream. In other words, the NC stream must have higher priority than the C stream,
and the C stream must have higher priority than j. This implies that the traffic stream considered, j, can also utilize the inter-
section while the NC stream interrupts the C stream that has higher priority than stream j. Then the form of the effective
capacity for this traffic stream is calculated as in Eq. (2).

Lj ¼ Sj  bj þ Sj  maxyhpNC  ð1  bj Þ ð2Þ
8hpNC

where yhpNC is the flow ratio of the higher priority NC traffic stream which is being taken advantage of. The flow ratio, y, is
defined as the ratio of the demand, Q, to the saturation flow, S (yhpNC ¼ Q hpNC =ShpNC ). The maximum of yhpNC is over all higher
priority NC traffic streams. Note that this methodology is presented for an under saturated intersection. Hence, the maxi-
mum of y is never expected to exceed 1. An example of this would be the car stream in Fig. 1. The conflicting pedestrian
 S.I. Guler, M. Menendez / International Journal of Transportation Science and Technology 5 (2016) 257–267 261

stream has priority over the car stream, however, the non-conflicting tram stream has priority over the pedestrian stream.
Therefore in this case, the car stream can take advantage of the higher ranked NC stream (i.e., the tram) when crossing the
pedestrians.

Step 2: Calculating the inputs (b) for the effective capacity equation

This step must be carried out for all C streams for the stream of interest, j.
For each stream of interest, j, the saturation flow is modified according to the flows of all its C streams. The idea is to
determine the bij value for every stream, i, conflicting with the stream of interest, j. To do so, the results of the empirical data
collection for the case when only two streams interact were used. As a result of analyzing this data, the bij values were deter-
mined for the interaction of different pairs of modes. At uncontrolled intersections with multiple traffic streams the effective
capacity for each individual traffic stream can be determined by considering every conflicting stream one by one and then
combining the effects together. This is a reasonable assumption since the individual interactions between streams would not
be affected by the presence of other streams (except for the case when the form of the capacity equation is as in Eq. (2)).
Fig. 3 shows the flowchart for determining the capacity modification factor b. The logic behind this figure is to consider all
streams, i, conflicting with the stream of interest, j, and determine whether they have priority or not.
By following the algorithm presented in Fig. 3, the resulting bij can be determined for every stream, i, conflicting with the
stream of interest, j. The final bj value for the stream of interest, j, would then be determined by Eq. (3) below:
Y
bj ¼ bij ð3Þ
8iconflicting with stream of interestj

Below is a description of the different equations mentioned in Fig. 3.

If a C stream, i, does not have priority over stream j, then:
bij ¼ 1 ð4Þ

In this case the saturation flow of j is not affected by the presence of the conflicting stream i. As an illustration, from Fig. 1
if the stream of interest, j, were the tram, and the C stream, i, were the pedestrian, the beta value for the tram would be 1.
On the other hand, if a C stream, i, has the same priority or higher priority over stream j then the bij value depends on the
mode of the C stream. These differences were identified as a result of the data collection of two stream interactions (Pitzinger
and Spacek, 2009).
If the conflicting stream i has the same priority as the stream of interest j, it is assumed that the capacity is modified
according to the ratio of both streams flow ratios, y. Notice that although not very common in the U.S., in uncontrolled inter-
sections in Europe it is very common to have multiple streams with the same priority. A common example would be two car
streams. They are normally served based on the order of arrival, or in some cases by yielding to the stream on the right side.
Then, one can expect that the flow share of the roadway between the two streams would depend on their flow ratios. The bij
in this case is given as in Eq. (5).
yi
bij ¼ ð5Þ
yi þ yj

If the conflicting stream i has higher priority than the stream of interest j, and it is composed by buses or trams, the bij is
given as in Eq. (6). As an illustration, from Fig. 1 if the stream of interest, j, were the pedestrian, and the C stream, i, were the
tram, the beta value for this conflict could be described by Eq. (6). Buses and trams have more regular flows, and only occupy

Stream i, conflicting with the

stream of interest, j

Stream i has lower priority Stream i has the same Stream i has higher priority
than stream j priority as stream j than stream j

Stream i:
Stream i: Stream i:
Pedestrian or Car at
Bus or Tram Car at roundabout
Intersection

Eq 4 Eq 5 Eq 6 Eq 7 Eq 8

Fig. 3. Flowchart for determining the capacity modification factor, b.

262 S.I. Guler, M. Menendez / International Journal of Transportation Science and Technology 5 (2016) 257–267

the roadway intermittently, hence, the saturation flow of stream j is only partially affected by their presence. Notice that yi
represents the flow ratio of the conflicting stream, i, hence, 1-yi can be thought of as the remaining percentage of time that
the roadway can be used by other traffic streams.

bij ¼ ð1  yi Þ ð6Þ

If the conflicting stream i has higher priority than the stream of interest j, and it is composed by pedestrians or cars at an
intersection, the bij is given as in Eq. (7). As an illustration, from Fig. 1 if the stream of interest, j, were the car, and the C
stream, i, were the pedestrian, the beta value for this conflict could be described by Eq. (7). Since cars and pedestrians have
more irregular flows and occupy the roadway typically more continuously, they affect the saturation flow of C streams to a
larger degree. This can be seen by the 1  yi term raised to the power of 3 in Eq. (7), which effectively means that stream j has
even less time available at the intersection. Recall that 1  yi > 0.

bij ¼ ð1  yi Þ3 ð7Þ

Also, cars at a roundabout are treated differently than cars at other intersections, as the priority rules are clearer, and the
merging behavior makes the system a bit more favorable towards cars entering the roundabout, compared to cars crossing a
normal intersection. Therefore, the bij in this case is given as in Eq. (8). Notice that the 1  yi term is raised to the power of 2
in this case, indicating a larger proportion of time is available to stream j when compared to Eq. (7).

bij ¼ ð1  yi Þ2 ð8Þ

Again, notice that Eqs. (58) were obtained based on empirical data collected in Switzerland. Please refer to Menendez
et al. (2015) for the details of this data collection. This reference has 5 tables that illustrate the accuracy of Eqs. (58) using
real world data. In general, the models can be used to predict delays of simple intersections within 20% for congested areas,
and within a second for uncongested areas. However, note that it is possible that the shape of these equations look different
for other countries with different driving behavior or traffic laws.
Notice too that these equations assume more or less uniform arrivals in the conflicting traffic stream i. Hence, the bij value
needs to be modified if there exists a signal upstream that changes the arrival pattern of the conflicting car stream. If the
conflicting stream’s arrival pattern is in platoons, this would imply a more regular flow pattern resulting in the remaining
capacity being better utilized by stream j. In this case, a new bij value can be calculated as:

1
bnew;ij ¼ bij  ð9Þ
ð1  yi £Þ
Q platoon;i
where £ ¼ Qi
defines the percentage of demand which arrives to the intersection in a platoon.

Step 3: Computing the final effective capacity of each stream

Combining the form of the capacity equation from step 1 (Eqs. (1) or (2)), and the beta value calculated in step 2 (Eq. (3)),
the final effective capacity, Lj, of each stream can be computed.

Step 4: Determining the delay

The equation to determine delay is adapted from the Highway Capacity Manual (2010) for the rest of this paper and is
given as in Eq. (10).
" rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi#
3600 8x
d¼  2 þ 900T ðx  1Þ þ ðx  1Þ2 þ ð10Þ
L TL

where d: average delay per vehicle [s/veh], L = effective capacity [veh/h], T = duration of analysis period [h], x = degree of uti-
lization defined as Q/L, and Q = demand [veh/h] (notice that x is slightly different than the flow ratio, y, since it is the ratio of
the demand to the effective capacity rather than the saturation flow). Note that the service time in the HCM equation has
been replaced with its equivalent of the headway (3600=LÞ minus the move-up time, and the move-up time has been
assumed to be 2 s as used in the HCM. Also, in the HCM formulation (meant for all-way stop-controlled intersections),
another 5 s are added to the delay defined above. This term is ignored in this analysis since when considering an uncon-
trolled intersection (contrary to an all-way stop-controlled intersection) it is possible to observe vehicles with no delays
at all. This can happen often in the intersections considered here, as vehicles do not have to stop unless there is a vehicle/-
pedestrian present in the conflicting direction. Notice that this is different than in most places in the U.S., where completely
uncontrolled intersections are not very common.
By following the methodology presented above, the vehicle delays at uncontrolled intersections can also be estimated.
However, the remainder of this paper will only focus on car delays.
 S.I. Guler, M. Menendez / International Journal of Transportation Science and Technology 5 (2016) 257–267 263

Empirical analysis of methodology

To test the methodology presented, empirical data were collected at two different locations and five intersections in
Zurich: a merge at Universitatsstrasse (Fig. 4a), and four legs of a multimodal roundabout in Letzigrund (Fig. 5).
Empirical data were collected at Universitatsstrasse during the morning peak hour on the 21st of August 2013. At this
intersection, multiple modes (cars, trams, buses and pedestrians) cross each other. The data was collected using 4 cameras
placed around the complex intersection. The exact multimodal interactions observed at this location, can be seen in Fig. 4a.
The video data was then manually extracted to obtain flows of cars for 3 different car streams (R1, R2, R3), 2 different pedes-
trian streams (P1, P2), a tram stream and a bus stream. The demand per hour for each stream (Q) can be seen within paren-
thesis in Fig. 4a. Also using this data, the empirical car delays were determined.
The second set of empirical data were collected at Letzigrund during the morning peak hour on the 29th of August, 2013
(Fig. 5). This location can be treated as four different complex intersections, where each leg approaching the large round-
about can be thought of as one of the intersections (see different shaded areas in Fig. 5). Notice that this is not a traditional
roundabout: it is very large and it has a tram line and stop that go through the middle of it; hence it is reasonable to treat

Tram R3
Bus (410)
(30)
(8)

P1
(58)
R1
(370)
P2
(94)
R2
(480)

(a)
Bus (5)

R3 (5)

Tram (1)

R1 (4)
P2 (2)
P1 (2)

R2 (3)

(b)
Fig. 4. (a) The car flows, tram, bus, and pedestrian crossing at Universitatstrasse; and (b) abstraction with priorities of each stream within parenthesis.
264 S.I. Guler, M. Menendez / International Journal of Transportation Science and Technology 5 (2016) 257–267

Fig. 5. The car flows, tram, bus, and pedestrian crossing at data collection site in Letzigrund.

each of the arms as a single, distinct intersection. To evaluate the 4 individual intersections, and also to determine how they
interact with each other, data was collected using 7 cameras placed around the tram stop. The data were then manually
extracted to obtain information on the demand and delay for the 8 car streams (R1 through R8); and the demand for the
13 pedestrian streams (P1 through P13) and the single tram stream (see Fig. 5). Notice that the pedestrian crossings on
the exit for each leg are not part of the analysis since these were found not to interact with the car streams considered.
The demand per hour (Q) of each stream considered can also be found in Fig. 5 in parenthesis.
To determine an origin–destination table for the roundabout, the approach leg, and the departure leg of each individual
car was determined. This can be seen in Table 2. Using this origin–destination table the car demands for each car stream can
also be determined.
Using the data described above and the methodology described in the previous section the delays for the car streams
were theoretically predicted and later compared with those empirically observed. To illustrate the process better, the

Table 2
Origin-destination table for Letzigrund (cars/h).

Origin
North West South East
Destination North 0 261 38 119
West 98 0 178 206
South 64 127 0 117
East 70 105 39 0
 S.I. Guler, M. Menendez / International Journal of Transportation Science and Technology 5 (2016) 257–267 265

analysis at Universitatsstrasse is described below, and the calculation of the L values for the three car streams, R1, R2, R3, is
shown step-by-step.
It was observed from the videos at Universitatsstrasse that for the duration of the data collection period, the stream R2
had priority, with cars coming from the R1 direction yielding to them (Fig. 4). The R2 direction was not affected by the trams
(as these two streams don’t intersect in reality), and hence the R1 direction could not take advantage of the tram stream (see
Fig. 4a). Also, even though the bus arrived to the intersection on the tram lane, it had to merge with cars to be able to com-
plete the left turn maneuver. Therefore, the bus and cars (R3) are assumed to have the same level of priority, as well as the
same influence on other traffic, and the same delay.
We started then by defining the abstraction of the intersection in order to determine the priority level and direction of
each traffic stream. This abstraction can be seen in Fig. 4b with the priority of each traffic stream shown within parenthesis
(lowest number implies highest priority). As can be seen, trams have first priority, pedestrians have second priority, and the
cars have the lowest priority at this intersection.
Then the four steps presented in the methodology section were followed to determine the effective capacity of the three
car streams.
Step 1: The flowchart presented in Fig. 2 is followed to determine the form of the equation for L for the three car streams.
R1 has a conflicting stream, which has higher priority (R2). However, its NC stream P2 has priority over R2. Therefore, R1
can take advantage of P2. The resulting form of the capacity equation is (based on Eq. (2)):
LR1 ¼ Scar  bR1 þ Scar  yP2  ð1  bR1 Þ
R2 has also a conflicting stream with higher priority, P2. However, it does not have any NC stream with higher priority
than the tram (i.e., there are no NC streams which R2 can take advantage of). Hence, the resulting form of the capacity equa-
tion is (based on Eq. (1)):
LR2 ¼ Scar  bR2
Since R3 has the lowest priority, both C streams, trams and R1, have priority over R3. However, the NC streams P1 and R2
have priority over R1. Hence, R3 can take advantage of these two modes. The resulting form of the capacity equation is
(based on Eq. (2)):
LR3 ¼ Scar  bR3 þ Scar  maxðyP1 ; yR2 Þ  ð1  bR3 Þ
Step 2: The flowchart presented in Fig. 3 is followed now to determine the b values for each car stream
The three C streams of R1 are P1, R2 and R3/Bus. P1 and R2 have priority over R1, resulting in bij values smaller than 1. On
the other hand, car stream R3/Bus does not have priority and hence has a bij value equal to 1. The resulting equation for the b
value for R1 is as follows (based on Eqs. (3), (4), and (7)):

bR1 ¼ ð1  yP1 Þ3  ð1  yR2 Þ3  1

The only C stream with higher priority than R2 is P2. Note that R2 does not need to cross the tram (see Fig. 4a). Therefore,
the resulting b value for R2 is as follows (based on Eq. (7)):

bR2 ¼ ð1  yP2 Þ3
All C streams have priority over R3 (Tram, R1 and P2). Therefore, all are a part of the b value for R3 as given in the equation
below (based on Eqs. (3), (6), and (7)):bR3 ¼ ð1  ytram Þ  ð1  yR1 Þ3  ð1  yP2 Þ3
There is no need to modify the capacity equations since there are no signals upstream.
Step 3: Putting together the results of steps 1, 2 and 3, the equations for calculating the three car streams’ L values are
found as

LR1 ¼ Scar  ½ð1  yP1 Þ3  ð1  yR2 Þ3  1 þ Scar  yP2  f1  ½ð1  yP1 Þ3  ð1  yR2 Þ3  1g

LR2 ¼ Scar  ð1  yP2 Þ3

LR3 ¼ Scar  ½ð1  ytram Þ  ð1  yR1 Þ3  ð1  yP2 Þ3  þ Scar  maxðyP1 ; yR2 Þ  f1  ½ð1  ytram Þ  ð1  yR1 Þ3  ð1  yP2 Þ3 g

Table 3
Empirical and calculated delays at Universitatstrasse.

Empirical Calculated Error

Priority Level Mode Demand S L Average delay
[Veh/h] [Veh/h] [s] [s] % [s]
3 R2 560 1750 1185 3.0 3.1 2% 0.1
4 R1 370 1650 604 13.1 13.3 1% 0.2
5 R3 410 1650 893 8.9 6.4 28% 2.5
266 S.I. Guler, M. Menendez / International Journal of Transportation Science and Technology 5 (2016) 257–267

Fig. 6. Empirical and calculated delays at the four intersections at Letzigrund.

Step 4: Finally, the delays of the car streams can be calculated using Eq. (10). Notice that Eq. (10) uses the degree of uti-
lization as an input, which is defined as the ratio of the demand to the effective capacity, L. Table 3 shows both the measured
and calculated average delay per car for the three car streams. The average measured delay has a low standard deviation,
typically less than 5% of the observed average delay.
As can be seen from Table 3, the methodology can predict the traffic streams’ average delays within 30%. The highest error
is observed for stream R3. From the video it was observed that some of the delay associated with this stream was also due to
cars travelling generally slower on this segment. Therefore, even if there were no conflicts at the intersection, some delay
was still measured. However, for all car streams the error in prediction is less than 3 s.
For the other four intersections, and using the methodology described above, the delays for the different car streams were
again estimated. The comparison of the calculated values to the empirically observed ones can be seen in Fig. 6.

Fig. 7. Scatter plot of empirical vs. calculated delay at Universitatstrasse and Letzigrund.
 S.I. Guler, M. Menendez / International Journal of Transportation Science and Technology 5 (2016) 257–267 267

A scatter plot of empirical vs. calculated delay is presented in Fig. 7. This data combines the results from the two locations
presented. As can be seen, the delay of all car streams is predicted within 3 s/vehicle of that empirically observed.

Conclusions

The goal of this paper was to propose a methodology to evaluate uncontrolled intersections where multiple transport
modes, or traffic streams, interact. The algorithm systematically checks for different possible scenarios and adapts the delay
equations for each traffic stream. These equations represent the percentage of time available to each mode to utilize the road
space. Data from five intersections in the city of Zurich was used to test the model. The results show that while the predic-
tions are not exact, typically the methodology does a good job of predicting the approaches which have larger delays. The
errors in average delay prediction are within 3 s/veh for the two locations observed, and often much smaller. While queue
lengths have not been mentioned in this paper, the same methodology as proposed in the HCM can be used to calculate
queue lengths. In the empirical evaluations, the errors when comparing empirical and calculated queue lengths were very
similar to the errors associated with average delays.
Overall, the algorithm presented is also beneficial since it is very systematic and can be easily programmed to automat-
ically determine the delays. The only inputs for this program would be the different traffic streams, their priorities, direc-
tions, saturation flows, and flows, which can all be easily observed. The methodology also allows for flexibility to adapt
to local conditions. The saturation flows can be easily measured for individual locations. Also, if necessary, the modifications
to the capacity used in step 2 can also be tested with data collection at intersections where 2 streams interact as was done in
Menendez et al. (2015). In any case, note that as with any calculation of this type, the results are just an approximation and
represent the overall magnitude of the delay, rather than the exact value.

Acknowledgement

Research contained within this paper was funded by the Research Commission for Road Transport in Switzerland (VSS) in
association with the project: Traffic flow at uncontrolled urban intersections with different modes of traffic (VSS 2011/308).
The authors are very grateful to the group of experts from VSS who provided invaluable feedback during the process;
especially to Dr. Peter Pitzinger and Prof. Peter Spacek, who established the bases for this methodology to be developed.

References

Brilon, W., Miltner, T., 2005. Capacity at intersections without traffic signals. Transport. Res. Rec. J. Transport. Res. Board 1920 (1), 32–40.
Brilon, W., Wu, N., 2001. Capacity at unsignalized intersections derived by conflict technique. Transport. Res. Rec. J. Transport. Res. Board 1776 (1), 82–90.
Bunker, J., Troutbeck, R., 2003. Prediction of minor stream delays at a limited priority freeway merge. Transport. Res. Part B Method. 37 (8), 719–735.
Gleue, A.W. 1972. Vereinfachtes verhfahren zur berechnung signalgeregelter knotenpunkte. Forschung Strassenbau und Strassenverkehrstechnik, 136,
Bonn.
Harders, J. 1968. The capacity of unsignalized urban intersection. Forschung Strassenbau und Strassenverkehrstechnik, 76, Bonn.
Li, H., Deng, W., 2008. Research on capacity at urban intersections without traffic signals. In: Proceedings of the First International Symposium on
Transportation and Development Innovative Best Practices. ASCE, Beijing, China, pp. 389–394.
Li, H., Deng, W., Tian, Z., Hu, P., 2009. Capacities of unsignalized intersections under mixed vehicular and nonmotorized traffic conditions. Transport. Res.
Rec. J. Transport. Res. Board 2130 (1), 129–137.
Li, H., Tian, Z., Deng, W., 2011. Capacity of TWSC intersection with multilane approaches. Proc. Soc. Behav. Sci. 16, 664–675.
Ma, D.F., Ma, X.L., Jin, S., Sun, F., Wang, D.H., 2013. Estimation of major stream delays with a limited priority merge. Can. J. Civ. Eng. 40 (12), 1227–1233.
Menendez, M., 2006. An analysis of HOV Lanes: Their Impact on Traffic (PhD thesis). Department of Civil and Environmental Engineering, University of
California, Berkeley, CA.
Menendez, M., Guler, I., Puffe, E. 2015. Traffic flow at uncontrolled urban intersections with attention to different modes of traffic; determination of
representative standard values and interrelations. Technical Report VSS 2011/308, Prepared for the Research Commission for Road Transport in
Switzerland, 2015.
Pitzinger, P., Spacek, P., 2009. Verkehrsqualität und Leistungsfähigkeit von komplexen ungesteuerten Knoten Analytisches Schätzverfahren. Eidgenössisches
Departement für Umwelt, Verkehr, Energie und Kommunikation UVEK, Bern, p. 1287.
Qu, Z., Duan, Y., Hu, H., Song, X., 2014. Capacity and delay estimation for roundabouts using conflict theory. Sci. World J. 2014,. http://dx.doi.org/10.1155/
2014/710938. 12 pages 710938.
Tanner, J.C., 1967. The capacity of an uncontrolled intersection. Biometrika 54 (3-4), 657–658.
TRB. Highway Capacity Manual. Transportation Research Board, National Research Council, Washington, D.C., USA, 2010.
Troutbeck, R., 1999. Capacity of limited-priority merge. Transport. Res. Rec. J. Transport. Res. Board 1678 (1), 269–276.
Troutbeck, R., Kako, S., 1999. Limited priority merge at unsignalized intersections. Transport. Res. Part A Policy Prac. 33 (3), 291–304.
Wu, N., 2000. Determination of capacity at all-way stop-controlled intersections. Transport. Res. Rec. J. Transport. Res. Board 1710 (1), 205–214.